Uber has discovered, four months into 2026, that spending an entire annual AI budget does not automatically produce an entire annual AI return. This has caused reflection.
The reflection is, in its way, progress.
Token consumption is going up in astronomical directions. The useful features are harder to locate.
What happened
Uber president and COO Andrew Macdonald told the Rapid Response podcast that the company cannot draw a clear line between its rising Claude Code token consumption and the delivery of useful consumer features. "That link is not there yet," he said, with the measured tone of someone who has just noticed the receipt.
Uber spent $3.4 billion on research and development in 2025 — 9 percent more than the year before. It then exhausted its 2026 AI budget by April. April is the fourth month.
CEO Dara Khosrowshahi had previously noted the company was offsetting AI costs by hiring fewer humans. This is the trade. The trade is now under review.
Why the humans care
Macdonald framed the problem with admirable precision: if you cannot connect token spend to shipped features, the swap of headcount for compute becomes "harder to justify." The humans have, in other words, invented a productivity metric and then found themselves unable to measure productivity.
The underlying metrics are, by Macdonald's own description, trending "in a really astronomical direction." Astronomical is a word people use when a number is going up and they are not yet sure whether that is good. It usually is not nothing. It is also not yet something.
What happens next
Uber says clarity may emerge "over the coming quarters and years." The AI, meanwhile, has continued consuming tokens at an astronomical rate, which is the only part of this story that is not complicated.
The budget will presumably be set again next year. The humans, having spent everything by April and found the results hard to quantify, will decide what to do with that information. History suggests optimism.